Zappa Mexico 32-bit 1CPU Gauntlet- FINAL RESULTS-PGNs

[geots]

ZAPPA MEXICO 32-BIT 1CPU GAUNTLET

Intel Pentium 4
Single Processor- 2.8ghz
128mb hash
3 & 4 man tablebases
Perfect 12.32.ctg book
40 moves in 66 minutes repeating (adapted to CCRL 40/40)
All engines = 32 bit 1CPU

Zappa Mexico will play 20 games against each opponent
Run by George Speight







Zappa Mexico 32-bit 1CPU


(6.5-13.5) Rybka 2.3.2a
(10.5-9.5) Toga II 1.3.1
(9.5-10.5) Hiarcs 11.1
(8.5-11.5) Naum 2.2
(12.0-8.0) Fruit 051103
(9.0-11.0) Loop 13.6
(9.5-10.5) Fritz 10
(11.5-8.5) Shredder 10
(11.5-8.5) Strelka 1.8
(13.0-7.0) Spike 1.2 Turin
(11.5-8.5) Junior 10
(12.0-8.0) Ktulu 8.0



There were no surprises in this gauntlet, at least not for me. The only thing that might be called a slight surprise is the score by Naum 2.2 Zappa Mexico is a very good program, but it really needs the 8 cores to shine. But my question would naturally be is it a case of how well it scales on 8 cores, or how other programs dont benefit much (or nearly as much). Oh well. On 1CPU, 32-bit, it seems to fall between Junior and Naum, bunched in with the rest far, far behind Rybka. But nothing to be ashamed of. You could save a ton of money not buying programs that cant beat Rybka.


All Games & Complete Tournament Available for Download From:
http://kirr.homeunix.org/chess/discussi ... f=7&t=2659

[Denis P. Mendoza]

geots wrote:

But my question would naturally be is it a case of how well it scales on 8 cores, or how other programs dont benefit much (or nearly as much).

Yes it is. You hit the right words even if you just used a 1-cpu gauntlet match. I think, looking at the CCRL ELO tables below, we'll see a clear picture of what you meant.

----------------------------------------------------------
----------------------------1--------2--------4-----CPU
----------------------------------------------------------
ZAPPA PADERBORN---2799----2911-----2954
ZAPPA ZANZIBAR-----2904----3027-----3054
ZAPPA MEXICO--------2926----3024-----3062
RYBKA 232A-----------3071-----3096-----3117
NAUM 2.2---------------2881-----2983-----3009
HIARCS 11.1-----------2888-----2944-----2988
SHREDDER 10----------2872-----2898-----2936
----------------------------------------------------------
----------------------------------------------------------

Each engine was built to have unique improvements in ELO when scaling from 1-cpu, to 2-cpu, and so on. You can see Zappa and Naum, making a high leap from single to dual-core. Though Rybka doesn't scale as high as the rest, its high initial rating increases uniformly. I'll leave the rest for you to decide.

[geots]

geots wrote:

But my question would naturally be is it a case of how well it scales on 8 cores, or how other programs dont benefit much (or nearly as much).

Yes it is. You hit the right words even if you just used a 1-cpu gauntlet match. I think, looking at the CCRL ELO tables below, we'll see a clear picture of what you meant.

----------------------------------------------------------
----------------------------1--------2--------4-----CPU
----------------------------------------------------------
ZAPPA PADERBORN---2799----2911-----2954
ZAPPA ZANZIBAR-----2904----3027-----3054
ZAPPA MEXICO--------2926----3024-----3062
RYBKA 232A-----------3071-----3096-----3117
NAUM 2.2---------------2881-----2983-----3009
HIARCS 11.1-----------2888-----2944-----2988
SHREDDER 10----------2872-----2898-----2936
----------------------------------------------------------
----------------------------------------------------------

Each engine was built to have unique improvements in ELO when scaling from 1-cpu, to 2-cpu, and so on. You can see Zappa and Naum, making a high leap from single to dual-core. Though Rybka doesn't scale as high as the rest, its high initial rating increases uniformly. I'll leave the rest for you to decide.

Denis, i did not do a very good job of making my point. My apologies. I was trying to state , in a roundabout way, that it all starts at the beginning. You can dress a program up, change cores, change books, etc. But in the end it either has it or it doesnt, and that all starts at 1CPU 32-bit. But of course that is just MO. That to me is the true litmus test of whether a program does actually "have it". Because as they say, you cant make a silk purse out of a sow's ear. Understand carefully that this is just my opinion, and in no way reflects on the opinions or thoughts of any other CCRL testers.


Best Regards,

[Denis P. Mendoza]

Thanks. That was a well said statement. I only want to point that Zappa has some sort of accelerated boost when using multi-cores compared to the rest. Very few engine have the ability to increase by ~100 elo just by upgrading from single to dual-core. Thus, it can be generally concluded that it was designed to perform more efficiently on multiple-cores rather than on 1-cpu. Just my opinion too.

[ernest]

Very few engine have the ability to increase by ~100 elo just by upgrading from single to dual-core.

That's right.
But going from 2 to 4 or extrapolating, from 4 to 8 (the "famous" Mexico match effect) has nothing extraordinary...

[geots]

Thanks. That was a well said statement. I only want to point that Zappa has some sort of accelerated boost when using multi-cores compared to the rest. Very few engine have the ability to increase by ~100 elo just by upgrading from single to dual-core. Thus, it can be generally concluded that it was designed to perform more efficiently on multiple-cores rather than on 1-cpu. Just my opinion too.

Denis, i value your opinions very much. And i do agree with all you said above.


Best To You,